Deploying Camera data model to Arduino 33 BLE sense

Feature requests
#1

I need to deploy my existing mask and no-mask trained model on edge implulse to Arduino 33 BLE Sense. But while downloading zip file I do not get a script for object detection. I just get a script for microphone and accelerometer. Is there a way out

#2

Hi @dhruvsheth12345,

The Arduino BLE Sense doesn’t have an image sensor, it is also too limited in RAM to run image classification.
A good alternative for image recognition is the OpenMV H7 Plus camera: https://docs.edgeimpulse.com/docs/openmv-cam-h7-plus

Aurelien

1 Like
#3

Dear Aurel, i was just reading this post because i’m not able to download a ‘simple’ bird classification model onto an Arduino BLE Sense with a little camera. Something like ‘will not fit in flash region’. Does it mean that the BLE cannot do any camera picture classification? Thanks a lot, Theo

btw it also happens compiling the camera example from Edge library download.

#4

Hi @ThPu,

Image classification (and FOMO) is now supported on the Arduino BLE Sense. However you’ll need to select a smaller model (ie: MobileNet v1 96x96 0.25 or smaller) so it can fit into the flash.

Aurelien

#5

Hi Aurel, aha thanks, i think i selected mobilenet v2 indeed. The same applies for the Nicla Vision which i’m trying out now, isn’t it?

Thanks a lot,

#6

Hello @ThPu,

The Nicla Vision is a bit more powerful and higher resources. But I’d suggest to start small and then increase the parameters as you reach the limits.

Best,

Louis

#7

Thanks Louis, gonna experiment with it.

#8

Hi Louis, the Nicla Vision works pretty well with a smaller model indeed. BTW do you have experience with using the WiFi on the Nicla, i’m intending to send predictions via WiFi to a MQTT broker. But can’t get the WiFi to work. Tx Theo

#9

Hello @ThPu,

I used the WIFI quite a lot on ESP32, including sending payloads to an MQTT broker but I not yet on the Nicla Vision.

Best,

Louis

#10

Oke thx Louis, i also used an Arduino Nano 33 IoT, and the WiFi/MQTT works fine. So was thinking maybe it’s the EdgeImpulse image that interferes with the WiFi, but it could also be an hardware issue or an Arduino library problem. Which ESP board as a substiture for the Nicla Vision would you recommend?

#11

I used extensively the ESP32-CAM AI Thinker but it’s not as powerful as the nicla vision.
I also love the OpenMV integration with the Nicla Vision/Portenta. It’s super easy to get something running in a couple of minutes including having a video return integrated in the IDE for debugging.

1 Like